Project Overview
Project Summary:
'Doctor Phish' is an AI-based voice phishing insurance. It uses AI that constantly learns call data to detect voice phishing. By aligning the interests of insurance stakeholders and encouraging contribution to the system, it ultimately aims for perfect prevention of voice phishing.
Identifying the Challenge
The Social Problem:
Due to the growing awareness among the public, the number of voice phishing incidents is decreasing, whereas the amount of loss per incident is increasing. New methods like 'Deep voice phishing' that utilize generative AI technology are anticipated to emerge as a serious social issue that is challenging to resolve by individuals alone.
Innovation and Uniqueness
Why Our Project Stands Out:
Current voice-phishing insurance policies are frequently overlooked due to their inadequate coverage. Current Voice-phishing detection AI models operate publicly without a sustainable business model. This approach limits the opportunity for user participation, which is crucial for enhancing model accuracy. Our AI-based insurance uses incentives like lower premiums to foster user participation, enhancing detection capabilities and offering greater coverage.
Insights and Development
Learning Journey:
Identifying the pain points of current voice-phishing detection and defining unique value propositions was challenging. Recognizing that true prevention exceeds mere alerts and requires user engagement, we merged AI with insurance for a positive loop. Also to overcome the dataset's limitations, we added call-center data, boosting our model's accuracy.
Development Process:
Our project initiated with leveraging the Hugging Face KoBERT model, fine-tuned for detecting voice phishing texts, alongside the OpenAI API for enhanced precision. Through collaborative efforts in Figma, designers, AI engineers, and product managers structured an efficient workflow. Reviewing existing services and brainstorming in meetings, we considered viable alternatives, leading to a well-integrated solution.
